Where possible, use java OO methods to write test harnesses by using or extending the "T3est" prefixed classes. There should be classes encapsulating each use case. Design for robustness to change.
See if optimised versions give identical results to non-optimised. (see the no3Opt flag in source code). See if behaviour is identical when switching on/off the precondition/postcondition/invariant/trace flags.
Consider reusing/creating tutorial scripts as test scripts. Remember
that the use cases provide the basis for a test script.